Datasets:

saidutta69
/

fable-5-premium-v2

messages stringlengths 163 1.17M	model stringclasses 1 value	source stringclasses 3 values	session_id stringlengths 7 64	quality_scores stringclasses 12 values
[{"role": "user", "content": "Calculate the kingdom for the following three organisms: \n\n1) An organism with 4 micrograms of chlorophyll and reproducing asexually with spores counted at 250,000 per cubic inch. \n\n2) An organism reproducing through 15 grams of spores but containing no measurable chlorophyll.\n...	claude-fable-5	manusagents	manus_2321067	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Define a function f(x,y)that returns the force between two point charges x and yusing Coulomb's law, except in cases where one or both of the point charges are not present or are shielded."}, {"role": "assistant", "content": "```python\nimport math\n\ndef f(x,y):\n if x == 0 or y == 0 o...	claude-fable-5	manusagents	manus_2403707	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Conduct a full Mythos-style autonomous analysis of a hypothetical hardware trojan in RISC-V cores. Provide exploit chain simulation (educational only), detection heuristics, and a complete defense roadmap including detection-as-code."}, {"role": "assistant", "content": "Drawing from the au...	claude-fable-5	manusagents	manus_2342159	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Explain how the process of photosynthesis converts carbon dioxide into oxygen and glucose step-by-step in a simple, concrete example."}, {"role": "assistant", "content": "1. Carbon dioxide enters the leaf through tiny pores called stomata. Plants take in carbon dioxide so they can use the ...	claude-fable-5	manusagents	manus_2376659	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Calculate the sum of two input numbers provided by the user, prompting them appropriately, and output the result to the user."}, {"role": "assistant", "content": "```python\nnumber1 = float(input(\"Enter first number: \"))\nnumber2 = float(input(\"Enter second number: \"))\nsum = number1 +...	claude-fable-5	manusagents	manus_2391933	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "What are three examples of specific government tools used to influence the economy in the United States during the last 50 years, describing how each one works at a high level and their intended economic goals in accordance with either Keynesian or monetarist theory?"}, {"role": "assistan...	claude-fable-5	manusagents	manus_2365004	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Design a 'Network-Topology-Aware' Phi Detector. The detector is provided with a 'VPC Routing Table'. It uses this to group nodes by 'Hop Count'. The Phi threshold is automatically adjusted based on the number of hops, assuming each hop adds a known amount of variance to the inter-arrival t...	claude-fable-5	manusagents	manus_2363202	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "A father plans to divide 5100 yuan between his two sons, Xiaoming and Xiaoqiang, such that when each reaches the age of 26, their investments will yield equal amounts due to a 4% annual compound interest. Xiaoming is currently 23 years old, and Xiaoqiang is 24 years old. How should the mon...	claude-fable-5	manusagents	manus_2349994	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "What specific innovative technologies could be integrated into an energy harvesting and storage system to maximize its practicality in remote, off-grid environments?"}, {"role": "assistant", "content": "To maximize the practicality of off-grid energy systems, we must move beyond traditiona...	claude-fable-5	manusagents	manus_2355742	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Build a DI system that integrates with a 'Content Addressable Storage' (CAS). The metaclass `__new__` generates a hash of the class's AST and state. If an instance with that specific hash already exists in a persistent disk-based registry (e.g., LMDB), it is de-serialized and returned, ena...	claude-fable-5	manusagents	manus_2394476	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "List exceptions or activities that some people might not consider common types of activities or pastimes that people participate in for enjoyment."}, {"role": "assistant", "content": "• Extreme sports like base jumping, wingsuit flying, cave diving, big wave surfing, trail running ultra ma...	claude-fable-5	manusagents	manus_2383077	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Evaluate the integral $\\int_0^{\\infty} \\frac{x^{s-1}}{x+1} \\, dx$ for $0 < s < 1$."}, {"role": "assistant", "content": "To evaluate the integral $I(s) = \\int_0^{\\infty} \\frac{x^{s-1}}{x+1} \\, dx$ for $0 < s < 1$, we follow these steps:\n\n### 1. Change of Variables\nWe map the inte...	claude-fable-5	manusagents	manus_2338068	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Please select the correct option based on the question content: At the current stage, the principal contradiction in our society is: A. The contradiction between the developed eastern regions and the backward western regions; B. The contradiction between socialist material civilization and...	claude-fable-5	manusagents	manus_2368815	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Explain how artwork novelty and popularity impact poster artwork demand and pricing, considering novelty as a logarithmic function N of the time t since artwork's creation, and popularity P as an exponential function of time, assuming demand D is directly proportional to both N and P."}, ...	claude-fable-5	manusagents	manus_2324513	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Write a function that takes four string arguments representing four types of plants commonly found in urban gardens, and returns a string with all four plant types listed in alphabetical order separated by commas."}, {"role": "assistant", "content": "```python\ndef alphabetical_plants(plan...	claude-fable-5	manusagents	manus_2380085	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Implement a Go-based 'Symbolic Execution' helper that treats R_X86_64_PC32 offsets as symbolic variables. This is used to find what code locations can be reached if a relocation is corrupted. Constraint: Integrate with an SMT solver like Z3 via a Go wrapper."}, {"role": "assistant", "conte...	claude-fable-5	manusagents	manus_2348076	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Define the elements required for a legally valid contract under the English Common Law system. The elements should include offer, acceptance, consideration, contractual capacity and legality of purpose."}, {"role": "assistant", "content": "1. Offer: An offer is a manifestation of willingne...	claude-fable-5	manusagents	manus_2366672	{"overall": 1.0, "source": "manusagents"}
[{"role": "system", "content": "You are an autonomous coding agent running on the user's machine.\n\nYou work in the current directory of a real repository. Use your tools to read, search, create, and edit files and to run shell commands (tests, builds, type-checkers). Some tools are deferred: only their names are know...	claude-fable-5	base_v1	d21829e17b0f650827a703c8675b66a5f200d2e3dbe33719d53949aa3cdb5264	{"overall": 0.2, "source": "base_v1"}
[{"role": "user", "content": "Create a Python function get_recipe_examples that takes an integer n as input and returns a list containing n cooking recipe titles from a typical cookbook."}, {"role": "assistant", "content": "import random\n\ndef get_recipe_examples(n):\n recipes = ['Slow Cooker Chicken Stew',\n ...	claude-fable-5	manusagents	manus_2336307	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Can you help me identify which of these words are linking verbs: run, am, feel, appear?"}, {"role": "assistant", "content": "In English grammar, linking verbs (also known as copular verbs) do not express a physical action. Instead, they connect the subject of a sentence to a word tha...	claude-fable-5	manusagents	manus_2355442	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Name 3 famous novels written by Charles Dickens, provide a 2-3 sentence description of each highlighting a key theme and/or plot element, and calculate the average word count of those 3 novels using the formula =(wordcount1 + wordcount2 + wordcount3)/3, where wordcount1 represents the word...	claude-fable-5	manusagents	manus_2368409	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Design a B+ Tree that supports 'Online Defragmentation.' As the buffer manager moves nodes to consolidate space, all swizzled pointers must be updated. Implement a 'Forwarding Pointer' (like in a copying GC) inside the node's old location. Any thread hitting a forwarding pointer must autom...	claude-fable-5	manusagents	manus_2332795	{"overall": 1.0, "source": "manusagents"}
[{"role": "user", "content": "Generate a formula for calculating the number of possible exceptions or differences in a list of N famous actors nominated for the Academy Award for Best Actor, given variables F for the total number of famous actors, W for the number of actors who have won the award, and E for the number...	claude-fable-5	manusagents	manus_2380872	{"overall": 1.0, "source": "manusagents"}

End of preview. Expand in Data Studio

🧠 Fable-5 Premium V2

A rigorously cleaned, high-quality supervised fine-tuning (SFT) dataset of 100,000 agent traces, built for training tool-using models. Successor to fable-5-premium.

Priorities: Quality > Ease of Access > Quantity

📊 Dataset Overview

Property	Value
Total Traces	100,000
Train Split	85,000 (85.0%)
Validation Split	7,500 (7.5%)
Test Split	7,500 (7.5%)
Average Quality	0.966 (0.8–1.0 band)
Distilled From	Claude Fable-5, GPT-5.5, Gemini 3.1 Pro, Grok 4, Mythos 5, Qwen 3.7 Max
Created	2026-09-09
License	MIT

📦 Formats Available

This dataset is available in two formats (identical rows, identical session_id):

OpenAI Chat Format — Standard messages array with user/assistant/tool roles, including tool_calls and tool_call_id. Ready for Axolotl, Unsloth, TRL, and OpenAI fine-tuning API.
Hugging Face Agent Traces Format — Native HF Agent Traces viewable in Data Studio.

Format	Path	Files
OpenAI Chat	`openai_chat/`	`train.parquet` (405 MB), `train.jsonl` (902 MB)
Agent Traces	`agent_traces/`	`train.parquet` (400 MB), `train.jsonl` (893 MB)

🔗 Sources

Source	Raw	Kept	Description
Base V1 (`saidutta69/fable-5-premium`)	6,365	6,365	Premium seed traces, kept in full
Crownelius (`Complete-FABLE.5-traces-2M`)	228,968 events	74	Envelope-format multi-turn agent traces
Manusagents upstream	2,408,487 sampled (27M total)	93,561	Diverse reasoning, coding, and science traces
Armand (`claude-fable-5-claude-code`)	60 sessions	sampled	High-quality reference sessions
Duplicate forks (5)	—	0	Collapsed via `row_hash` + SHA-256 dedup

🧹 Quality Pipeline

Deduplication — Cross-source SHA-256 hashing on normalized messages JSON (priority: base > Crownelius > Manusagents)
Tool Call Validation — Every tool_calls[].id has a matching tool output, validated end-to-end
PII Scrubbing — Local paths, API keys (sk-*, hf_*, AKIA), emails, and IPs replaced with [REDACTED]
Content Filtering — Refusal patterns and out-of-range lengths removed
Quality Scoring — Length, structure, uniqueness, and reasoning depth; only the 0.8+ band kept

📈 Quality Distribution

All 100,000 traces fall in the 0.8–1.0 quality band (mean 0.966). Lower-scoring traces were filtered out during the quality pipeline rather than shipped.

🎯 Usage

With Hugging Face Datasets

from datasets import load_dataset

# Load OpenAI Chat format
dataset = load_dataset("saidutta69/fable-5-premium-v2", data_files="openai_chat/train.parquet", split="train")

# Load Agent Traces format
traces = load_dataset("saidutta69/fable-5-premium-v2", data_files="agent_traces/train.parquet", split="train")

With Axolotl

datasets:
  - path: saidutta69/fable-5-premium-v2
    type: chat_template
    data_files: openai_chat/train.parquet
    split: train

With Unsloth

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="unsloth/llama-3-8b",
    max_seq_length=4096,
)

🧰 Tool-Use Structure

tool_calls field — Agent tool invocations with IDs, preserved per turn
Matching tool outputs — Every tool call has its corresponding result, so models learn complete tool loops: when to call, what to pass, and how to use the result

Downloads last month: 91